学术报告201609期

报告题目：Dense  Prediction on Images using Very Deep Convolutional Networks

报 告 人：Chunhua  Shen（沈春华） (Professor  at School of Computer Science, The University of Adelaide; Australian Research  Council (ARC) Future Fellowship; Project Leader and Chief Investigator at the  Australian Research Council Centre of Excellence for Robotic Vision  (ACRV))

报告时间：2016年4月23日（星期六） 下午3:00

报告地点：南一楼中311室

邀 请 方：“多谱信息处理技术”国家级重点实验室

Abstract:  In this talk I  will present an overview of my recent results on deep  learning.

First I will  introduce two deep structured learning methods. Structured output learning  concerns the problem of predicting multiple variables that have dependency, with  Conditional random field (CRF) as a typical example. The first application is to  learn depth from single monocular images using a deep structured learning  scheme. The unary and pairwise potentials of continuous CRFs are learned in a  unified deep CNN framework. For the second application, a new, efficient deep  structured model learning scheme is proposed for semantic segmentation. We  achieve best reported results on seven public benchmark  datasets.

Inspired by the  deep residual network, originally designed for classification, we design very  deep fully convolutional networks (FCNN) which significantly improve performance  on dense pixel-level prediction for both high-level and low-level problems,  including semantic segmentation, depth estimation, denoising,  super-resolution.

报告人简历：Chunhua  Shen（沈春华）is  a Professor at School of Computer Science, The University of Adelaide. From 2012  to 2016, he holds an Australian Research Council (ARC) Future Fellowship.  He is a Project Leader (machine learning for robot vision) and Chief  Investigator (one of I3 CIs) at the Australian Research Council Centre of  Excellence for Robotic Vision (ACRV), which reserves a total of $ 20  million in federal funding between 2014 and 2020. He is also involved in the  Data to Decisions CRC Centre (D2DCRC), in particular on the projects of large  scale image classification and text analysis. Before he moved to Adelaide in  2011, he was with the computer vision program at NICTA (National ICT Australia),  Canberra Research Laboratory for about six years, working with Professor Richard  Hartley. His research interests are in the intersection of computer vision and  statistical machine learning. Recent work has been on deep learning with  application to street scene understanding. He has recently published 17 papers  on JMLR, TPAMI and IJCV, and about 50 papers on  NIPS, ICML, CVPR, ICCV and ECCV. His  publications have been cited for more than 3200 times (Google Scholar). He is  serving as the Associate Editor of IEEE Transactions on Neural Networks and  Learning Systems, and the Program Members of CVPR, ICCV and  ECCV.